Reddit user calls out AI influencer videos conflating RAM and VRAM for local model runs
forevergeeks · reddit · 2026-10-07
A Reddit poster argues that influencer videos about running local models all use the same script: identical benchmarks, kitchen demos, and terminology. The core complaint: videos claiming Qwen3.8-Flash-Next runs on "12GB of RAM" via a new runtime called Strata actually need 12GB of VRAM (a dedicated GPU), plus roughly 64GB of system RAM and 80GB of SSD — a distinction that misleads newcomers.
More from Fun
- 16-year-old rescued from BC's 'Widowmaker' mountain after letting Claude plan his hike — Polymarket · 2026-10-07
- 'Insane alpha in being a normal person right now' becomes AI circle's favorite quip — kyliebytes · 2026-10-07
- Carmack on seeing mirror balls and remembering OpenGL sphere maps and light probes — ID_AA_Carmack · 2026-10-07
- AI founder named Aristotle Socrates is too on-the-nose for Twitter — var_epsilon · 2026-10-07
- Jason Wei writes a local's 3-hour San Francisco walking guide for COLM visitors — _jasonwei · 2026-10-07
- GLM 5.3's Quirk: It Keeps Saying "None" to Itself While Working — lucasmeijer · 2026-10-07